ABSTRACT
This study argues for statistical modeling of phonetic data in providing a phonological representation of tones using Chao’s letters. It first focuses on phonetic examinations of several phonetic cues subject to a perceptual experiment. Then, the perceptual study shows that other cues found in the phonetic examination do not contribute significantly to the discrimination of allotone pairs after voiced versus voiceless onset, and that F0 contours are sufficient in discriminating those allotone pairs without onset consonants. Finally, the F0 contours are statistically modeled, and the underlying pitch targets statistically tested to be quadratic correspond relatively well to record in the fieldwork. The fitted values obtained from the optimal model were calculated, and sample quantiles were obtained. The adjusted onset, the turning point (if quadratic), and the offset were transformed into Chao’s five points by evaluating the values against the proposed quantiles. The final representations provide similar basic tonal shapes as described in the fieldwork, with some differences in the exact integers used for the onset, turning point, and offset. This method provides a representation more consistent with normalized phonetic F0 values, taking the perceptual aspect into consideration.
